2022年關山－池上地震誘發之同震山崩研究

Abstract

同震山崩（coseismic landslide），指由地震直接誘發的山崩，是地震發生時最主要致使人員傷亡的其中一種災害。若能系統性地整合同震山崩的崩塌位置和崩塌面積等資訊，建立完整且準確的山崩目錄（landslide inventory），再進一步將山崩目錄與地震、地形和地質因子進行相關性分析，將有助於理解同震山崩的分布特徵與發生機制，進而提升災害應變與預警能力，減低傷亡以及經濟損失。 2022年9月17日與18日，臺東縣關山鎮與池上鄉相繼發生芮氏規模6.6和6.8的關山－池上地震序列，引發臺灣東部縱谷南段一帶相當多的同震山崩。由於目前公開的研究資料似乎仍尚未完整，因此本研究透過比較地震前後高解析度的Pléiades衛星影像，先利用常態化差值植生指標（NDVI）快速辨識山崩區域，再以人工視覺判讀圈繪山崩範圍，完成建立2022年關山－池上地震之同震山崩目錄，其中總共圈繪876個山崩，總崩塌面積約1.9×106平方公尺。 後續分析同震山崩與各項因子的關聯性顯示，震度、震央距離、斷層距離以及與河流距離因子皆大致呈現正相關的趨勢；大部分同震山崩發生在PGA超過250 gal、PGV大於12 cm/s的區域；山崩崩壞比以火成岩區域最高，坡度則集中於45度至60度，坡向集中於南向的坡面，高程則集中在500至1000公尺的範圍。分析結果顯示地震因子（震度、震央距離、斷層距離、PGA、PGV）與岩性是控制同震山崩分布最為重要的因子，且坡度、坡向等其他地形因子亦同時影響同震山崩的空間分布。 本研究進一步與前人研究結果進行比較，發現山崩目錄在涵蓋少數大面積的山崩時，即可大致反映出同震山崩的空間分布，但在部分因子的統計上仍會因為山崩目錄不完整而產生差異，顯示建立完整山崩目錄的必要性。最後，透過比較本研究與2024年花蓮地震的同震山崩案例，顯示縱谷南段區域的岩性特性相較於花蓮地區更為脆弱，使關山－池上地震同震山崩具有較低的PGV閾值，且更集中於相對平緩的坡度與低海拔的區域，再次顯示岩性與其特性是臺灣東部同震山崩空間分布與邊坡穩定度的關鍵因子。本研究之成果不僅釐清各項因子之間的相互關係，並為花東縱谷南段提供同震山崩潛勢評估與防災應用的參考。

Coseismic landslides, defined as landslides triggered by earthquakes, are major hazards during seismic events and often result in considerable casualties. Systematically establishing an accurate landslide inventory that includes both the location and area of the landslides, followed by analysis of the correlation between landslide occurrence and seismic, topographic, and geological factors, is crucial for understanding the spatial distribution and mechanisms of coseismic landslides. Such efforts can help delineate potential landslide zones, thereby enhancing disaster response, improving early warning capabilities, and mitigating hazards. A major earthquake sequence occurred in Taitung, Taiwan, on 17 and 18 September 2022, with a ML 6.6 foreshock in Guanshan followed by a ML 6.8 mainshock in Chihshang. The Guanshan-Chihshang earthquake sequence triggered many coseismic landslides in the southern Longitudinal Valley, eastern Taiwan. Since previous landslide inventories for this event were mostly incomplete, this study utilized high-resolution pre- and post-event Pléiades satellite images to establish a comprehensive coseismic landslide inventory for this event. The Normalized Difference Vegetation Index (NDVI) was initially employed for rapid detection of landslide areas, followed by manual visual interpretation to establish the coseismic landslide inventory for the 2022 Guanshan-Chihshang earthquake. Overall, this study identified a total of 876 landslides, with a combined area of approximately 1.9×106 m2. A subsequent analysis of landslide controlling factors was conducted. The results revealed a positive correlation between landslide frequency and four factors, including seismic intensity, epicentral distance, distance to faults, and distance to rivers. Most coseismic landslides occurred in areas where the peak ground acceleration (PGA) exceeded 250 gal and the peak ground velocity (PGV) surpassed 12 cm/s. In terms of landslide area ratio, the highest ratio occurred in igneous rock regions, on slopes from 45° to 60°, on south-facing slopes, and within an elevation range of 500 to 1000 meters. These results indicate that seismic factors (intensity, epicentral distance, distance to faults, PGA, and PGV) and lithology are the most critical controls governing the spatial distribution of coseismic landslides, while topographic factors such as slope gradient and slope aspect concurrently influence the landslide distribution. Furthermore, a comparative analysis with previous studies showed that although an inventory capturing a few large-scale landslides can broadly reflect the spatial distribution of coseismic landslides, discrepancies still arise in certain factors due to inventory incompleteness. This underscores the importance of a complete landslide inventory. Finally, comparing the results of this study with the coseismic landslide cases of the 2024 Hualien earthquake, it is revealed that lithological characteristics in the southern section of the Longitudinal Valley is appear to be more fragile than those of the Hualien region. This produced a lower PGV threshold for landslide initiation during the Guanshan–Chihshang earthquake, with failures concentrated on relatively gentler slopes and at lower elevations. This comparison emphasizes that lithology is the dominant factor controlling slope stability and the spatial distribution of coseismic landslides in eastern Taiwan. The results of this study not only clarifies the relationships between various landslide controlling factors, but also provide an important basis for coseismic landslide susceptibility assessment and disaster mitigation applications in the southern Longitudinal Valley.